Automated model-based lesion tracking in CT: colorectal liver metastases as a use case for development and...

Area of Science:

• Medical Imaging
• Artificial Intelligence in Radiology
• Oncology

Background:

• Manual tracking of colorectal liver metastases (CRLM) on CT scans is time-consuming and prone to variability.
• Accurate longitudinal assessment of CRLM is crucial for therapy response evaluation.

Purpose of the Study:

• To develop and evaluate an automated CT liver lesion-tracking algorithm (Auto-MBT) for CRLM.
• To assess the algorithm's ability to match lesions over time, detect new metastases, and provide per-lesion confidence scores.

Main Methods:

• Developed a machine learning-driven, automated model-based lesion tracking (Auto-MBT) algorithm.
• Compared Auto-MBT performance with deformable registration plus overlap and deformable registration plus Auto-MBT methods.
• Evaluated performance on a dataset of 87 adults with unresectable CRLM and an external melanoma dataset, stratifying by lesion size and count.

Main Results:

• Affine registration + Auto-MBT achieved high performance in matching (F1=0.989), detecting new lesions (90%/93%), and identifying disappeared lesions (93%/96%) on the CRLM dataset.
• The algorithm outperformed other tested methods, including deformable registration + overlap (F1=0.797).
• Per-lesion confidence scores differentiated accepted from rejected matches, indicating triage utility.

Conclusions:

• Auto-MBT accurately tracks CRLM, including new and subcentimeter lesions, improving longitudinal assessment.
• The algorithm provides per-lesion confidence scores to triage, not replace, clinician review for therapy response assessment.
• Automated total-tumor tracking complements RECIST criteria by quantifying all lesions.